Educational AI systems have historically focused on evaluating learning outcomes, most commonly through correctness‑based metrics and numerical scores. While effective for assessment and grading, such systems provide limited insight into how students are learning and offer insufficient support for instructional decision‑making. This white paper presents a Teaching‑Oriented Diagnostic Methodology designed to bridge the structural gap between assessment results and teaching actions. By integrating response accuracy, relative response time, and instructional attribution, the methodology infers to learn states that are interpretable, actionable, and ethically constrained. Response time is employed strictly as a relative behavioral signal, never exposed to learners and never used for evaluation or ranking. Its sole purpose is to help distinguish stable mastery from effortful correctness within instructional contexts. Diagnostic outputs are expressed in pedagogical language, enabling teachers to adjust instruction while retaining full professional responsibility. The framework adopts a constrained four‑quadrant diagnostic structure that prioritizes interpretability, reproducibility, and system governance over predictive complexity. To ensure long‑term reliability, it incorporates simulation‑based validation and regression testing mechanisms, preventing semantic drift and misuse as systems evolve. This methodology does not aim to measure learner ability, diagnose psychological traits, or predict future performance. Instead, it supports responsible, teacher‑centered instructional decisions, providing a practical foundation for trustworthy educational AI. For v2.0, updated Appendix B and Appendix C.
Jianbo Li (Sun,) studied this question.